20 models · on the curve since 2018

OpenAI

Value position · the site’s read Output layer

OpenAI’s scarcity sits on the output layer, the part of the stack that commoditises fastest. Its durable bet is ChatGPT as a daily habit and the platform that routes work to whichever model wins. The frontier model itself is the commodity.

What would change this readChatGPT loses its consumer default to a rival with better distribution, or frontier quality stops being matchable by the next lab within months.

Recent cadence
6 in last 12 mo
6 the year before
Landmark density
9 / 20
defining releases
Weights
Closed / API
GPT-5.6 flagship

Founded 2015San Francisco, USA

The substrate underneath figures as of Aug 2026

Trains and serves on

NVIDIA GPUs carry the training fleet and the majority of the inference stack. Around that: clouds from Microsoft, Oracle, AWS, CoreWeave and Google Cloud, and silicon from NVIDIA, AMD, AWS Trainium, Cerebras and an own-design chip co-developed with Broadcom.

ConcentrationThe portfolio spans five clouds and five silicon vendors, yet OpenAI states that NVIDIA still carries training and most inference — so one chip vendor sits under most of the compute.

Flagship, per million tokens

$5 in $30 out

GPT-5.6 Sol, the named flagship

Was

$1.25 / $10

GPT-5 · Aug 2025

Weights
Partial. gpt-oss-120b and gpt-oss-20b under Apache 2.0; every GPT-5.x frontier model is closed.
Where you buy it
Own API · ChatGPT and Codex · Azure AI Foundry · AWS Bedrock (to GPT-5.5)
Structure
The nonprofit OpenAI Foundation controls OpenAI Group PBC, a Delaware public benefit corporation, through special voting rights that let it appoint and replace any director.
Raised
$122B committed at the March 2026 round, anchored by Amazon, NVIDIA and SoftBank
Valuation
$852B post-money (31 Mar 2026)
Revenue
$2B per month, about $24B annualised (31 Mar 2026)

Biggest single dependencyThe training fleet and most of the inference stack run on NVIDIA GPUs, making NVIDIA supply the single largest physical dependency.

Revenue is the last figure OpenAI confirmed, in March 2026; later ARR numbers in circulation are estimates. Total cumulative funding and headcount are not published and third-party trackers disagree by roughly two-fold, so neither appears here. The flagship list price has risen while the cheapest tier collapsed to $0.20/$1.20 — the fall is happening below the frontier, not at it.

On the durability curve frontier era
2023202420252026 ChatGPTGPT-5.6 (Sol / Terra / Luna) ARCHITECTURE the substrate · compounds  → OUTPUT per model · commoditises  → CROSSOVER the durability thesis  → DURABLE VALUE-CAPTURE ↑ the margin a release keeps · not raw capability TIME · FRONTIER MODEL RELEASES →
Output · value a model keeps Architecture · the compounding substrate OpenAI release

Hover any point to read the release. The y-axis is qualitative; only the x-axis carries dates.

When to reach for them practical read
Broadest reach
Reach for it when you want the widest consumer and developer footprint (ChatGPT, Nov 2022).
One unified system
Pick GPT-5 when you want the router to choose the model tier for you (Aug 2025).
Hard reasoning
Use the o-series for inference-time reasoning on difficult problems (o1, Sep 2024).
The landmark releases 9 of 20 models
  1. Jul 2026 GPT-5.6 (Sol / Terra / Luna) Latest Three price-performance tiers, GA across ChatGPT, Codex and the API FlagshipCurrentTiered
  2. Apr 2026 GPT-5.5 Flagship system of the 5.5 line Flagship
  3. Aug 2025 GPT-5 Unified system routing across model tiers UnifiedFlagship
  4. Sep 2024 o1-preview First inference-time reasoning model ReasoningFirst of kind
  5. May 2024 GPT-4o Omni real-time text, vision and audio OmniMultimodal
  6. Mar 2023 GPT-4 Multimodal flagship, large capability jump MultimodalFlagship
  7. Nov 2022 ChatGPT Consumer chatbot — fastest app to 100M users Consumer launch
  8. Jun 2020 GPT-3 175B-parameter model behind a public API 175BAPI
  9. Jun 2018 GPT-1 First generative pretrained transformer First GPTResearch
Full release logall 20, newest first
  1. 2026
  2. Jul GPT-5.6 (Sol / Terra / Luna) ◆ Three price-performance tiers, GA across ChatGPT, Codex and the API FlagshipCurrentTiered
  3. Apr GPT-5.5 ◆ Flagship system of the 5.5 line Flagship
  4. Mar GPT-5.4 Point release Point release
  5. Feb GPT-5.3-Codex Coding-focused Codex variant CodingAgentic
  6. 2025
  7. Dec GPT-5.2 Point release Point release
  8. Nov GPT-5.1 Iterative quality and steerability upgrade Upgrade
  9. Aug GPT-5 ◆ Unified system routing across model tiers UnifiedFlagship
  10. Apr o3 / o4-mini Full reasoning lineup with tool use ReasoningTool use
  11. Apr GPT-4.1 API-first coding and instruction model CodingAPI
  12. Feb GPT-4.5 "Orion" — large non-reasoning research preview Research preview
  13. Jan o3-mini Efficient reasoning at lower cost ReasoningEfficient
  14. 2024
  15. Sep o1-preview ◆ First inference-time reasoning model ReasoningFirst of kind
  16. May GPT-4o ◆ Omni real-time text, vision and audio OmniMultimodal
  17. 2023
  18. Nov GPT-4 Turbo Cheaper, faster, longer context 128K ctxCheaper
  19. Mar GPT-4 ◆ Multimodal flagship, large capability jump MultimodalFlagship
  20. 2022
  21. Nov ChatGPT ◆ Consumer chatbot — fastest app to 100M users Consumer launch
  22. Mar GPT-3.5 InstructGPT-tuned for following instructions Instruct-tuned
  23. 2020
  24. Jun GPT-3 ◆ 175B-parameter model behind a public API 175BAPI
  25. 2019
  26. Feb GPT-2 Staged release over 2019 on safety concerns Staged release
  27. 2018
  28. Jun GPT-1 ◆ First generative pretrained transformer First GPTResearch
Where they sit on weights versus the field
Open weights
DeepSeek Moonshot AI Zhipu (Z.ai)
Mixed
Google Meta Mistral AI Alibaba (Qwen)
Closed / API
OpenAI Anthropic xAI
The durability take on OpenAI from the archive
The Model Is Not the Moat If frontier capability keeps centralising, the durable edge shifts outward into trust, workflow fit, and the surrounding package. Read → You're Not Comparing Models. You're Comparing Contracts. Agent benchmarks don't measure models. They measure contracts. Two teams running the same model can publish different scores, and both can be honest. Read → The Forest Floor Is The Product Most people are optimising for canopy. The work that survives the next five AI releases is built from the forest floor up. Read →
About OpenAI at a glance
Founded
2015
HQ
San Francisco, USA
Leadership
Sam Altman (CEO)
Backers
Microsoft
Flagship
GPT-5.6
Weights
Closed / API
Also builds
ChatGPT, Sora, API platform
Known for
ChatGPT, GPT series

Founded in December 2015 by Sam Altman, Greg Brockman, Ilya Sutskever, Elon Musk and others as a nonprofit lab, later restructured as capped-profit. Headquartered in San Francisco, it is best known for the GPT series and ChatGPT, run as closed-weight models behind an API and consumer apps.

The frontier chart → Read the archive →